Remove SynthID watermarks from images
SynthID is an invisible watermark embedded directly into the pixels of AI-generated images. RAIW uses a two-stage diffusion pipeline to disrupt the hidden signal while preserving the image's structure, then runs a dedicated face-restoration pass.
Quick clean is free with Standard output up to 12 MP. Keeping the original resolution above that size requires paid processing.
What we have actually verified
The current production profile is pinned to a deterministic seed. Its release-candidate outputs were checked with the matching OpenAI and Gemini provider verifiers and those exact files no longer reported the watermark. That evidence applies to the tested files and seed, not to every image or future watermark revision.
- Tested condition
- Pinned production pipeline and seed
- Verification
- Matching OpenAI and Gemini provider tools
- Limit
- No universal guarantee against future detectors
The invisible marker problem
SynthID is an invisible watermarking technology used by Google and OpenAI. Unlike a visible logo, you cannot see it. It is a machine-readable signal embedded in the image pixels, designed to survive common edits like cropping, resizing, and compression.
It is applied automatically to images generated by Google Gemini and Imagen, as well as ChatGPT, Codex, and the OpenAI API, allowing detection tools and platforms to identify their origin.
How the removal process works
RAIW removes these hidden markers by rebuilding the image with a two-stage diffusion pipeline. The first stage uses edge guidance to preserve composition and structure while disrupting SynthID v1 frequency patterns and SynthID v2 pixel-level noise. The second stage detects faces in the original image and restores those regions separately to reduce identity drift. The result can still differ from the original, especially in fine text and facial details, and no tool can guarantee removal against a closed, evolving detector.
Because it operates on the underlying image structure, this approach can also disrupt other hidden watermarks like StableSignature and TreeRing.
SynthID is updated over time, so removal is a moving target. We keep the diffusion pass calibrated to the versions in circulation, but the newest watermark revision may not be fully removed until we re-tune - if a detector still flags your image afterward, it likely carries a version we have not caught up to yet.
Processing and privacy
Removing hidden markers requires heavy computation (diffusion), so this specific mode runs on our secure cloud GPUs.
- Cloud GPU: Processing can take several minutes, including model startup and inference.
- Cost: From $1.49 per image.
- Privacy: Uploaded images and results currently have no automatic expiry. We retain them until they are no longer needed or we receive a valid deletion request that identifies them. We never sell them.
Related tools
Remove visible and hidden watermarks
Use the All-in-one mode to remove visible, invisible, and metadata signals.
Remove all AI metadata
Clean EXIF and XMP tags without altering pixels.
Remove the Gemini watermark
Remove visible sparkle overlays and hidden markers.
Remove C2PA Content Credentials
Strip provenance data often embedded with SynthID.
Remove the Made with AI label
Clean common metadata that triggers platform disclosures.
Frequently asked questions
What does invisible watermark removal mean?
Will the image quality change?
Does it remove the very latest SynthID?
Is it enough to just remove the visible watermark?
Can I just crop the image to remove SynthID?
Invisible watermark removal starts at $1.49 per image, with no account required.